Since characteristics other than gender are likely to be useful for adjustment, [TABULAR DATA FOR TABLE 4 OMITTED] we reviewed which correlated demographics in the CNIDS interview - age, gender, education, and income - would also pass the consistency test and would be comparable in the CNIDS and the 1995 Census data.
The CNIDS raw data, when weighted using the Project 2000 weights, has three properties:
One additional factor that may contribute to the statistical problems with the CNIDS weighted sample, besides a deficient weighting scheme, is the cooperation rate, calculated as the percentage of eligible individuals who agreed to participate in the survey.
The sample cooperation rate for the CNIDS was not publicly released but is reported in the proprietary Final Report , available for a fee from the CommerceNet Consortium or from Nielsen Media Research.
The CNIDS estimates of Internet and Web use are logically inconsistent because individuals saying they used the Internet in one question only to later say they did not use the Internet were included in counts of use.
The effect of reweighting the CNIDS raw data is shown by comparing columns (b) and (c) in Table 2.
For example, column (b) shows that CNIDS estimates 16.9 million Americans use the Web, while the Project 2000 estimate in column (c) is that 14.1 million Americans use the Web.